Hotfix for case where there are no centroids detected in inference.

sleap/nn/inference.py

@@ -100,6 +100,7 @@ def predict_chunk(self, img_chunk, chunk_ind, chunk_size):
         """Runs the inference components of pipeline for a chunk."""
 
         if "centroid" in self.policies:
+            # Detect centroids and pull out region proposals.
             centroid_predictor = self.policies["centroid"]
             region_proposal_extractor = self.policies["region"]
 
@@ -114,21 +115,26 @@ def predict_chunk(self, img_chunk, chunk_ind, chunk_size):
                 region_proposal.RegionProposalSet.from_uncropped(images=img_chunk)
             ]
 
-        # for region_ind in range(len(region_proposal_sets)):
-        #     region_proposal_sets[region_ind].sample_inds += chunk_ind * chunk_size
-
         if "paf" not in self.policies:
+            # If we don't have PAFs, we must be doing topdown or single-instance.
             if "topdown" in self.policies:
                 topdown_peak_finder = self.policies["topdown"]
             else:
                 topdown_peak_finder = self.policies["confmap"]
 
+            if len(region_proposal_sets) == 0:
+                # No region proposals were found, so just return empty result.
+                return defaultdict(list)
+
+            # Only one scale region proposals in topdown inference.
             rps = region_proposal_sets[0]
 
+            # Find peaks in each sample of the region proposal set.
             sample_peak_pts, sample_peak_vals = topdown_peak_finder.predict_rps(rps)
             sample_peak_pts = sample_peak_pts.to_tensor().numpy()
             sample_peak_vals = sample_peak_vals.to_tensor().numpy()
 
+            # Gather instances across all region proposals for this chunk.
             predicted_instances_chunk = topdown.make_sample_grouped_predicted_instances(
                 sample_peak_pts,
                 sample_peak_vals,
@@ -140,16 +146,19 @@ def predict_chunk(self, img_chunk, chunk_ind, chunk_size):
             confmap_peak_finder = self.policies["confmap"]
             paf_grouper = self.policies["paf"]
 
+            # Find multi-instance peaks in each region proposal set.
             region_peak_sets = []
             for rps in region_proposal_sets:
                 region_peaks = confmap_peak_finder.predict_rps(rps)
                 region_peak_sets.append(region_peaks)
 
+            # Group peaks into instances using PAFs.
             region_instance_sets = []
             for rps, region_peaks in zip(region_proposal_sets, region_peak_sets):
                 region_instances = paf_grouper.predict_rps(rps, region_peaks)
                 region_instance_sets.append(region_instances)
 
+            # Gather instances across all region proposals for this chunk.
             predicted_instances_chunk = defaultdict(list)
             for region_instance_set in region_instance_sets:
                 for sample, region_instances in region_instance_set.items():

sleap/nn/region_proposal.py

@@ -467,6 +467,10 @@ def extract_region_proposal_sets(
             sample_inds = size_grouped_sample_inds[box_size]
             bboxes = size_grouped_bboxes[box_size]
 
+            if len(bboxes) == 0:
+                # Skip region proposal set if we found no bboxes.
+                continue
+
             # Extract image patches for all regions in the set.
             patches = extract_patches(
                 imgs, tf.cast(bboxes, tf.float32), tf.cast(sample_inds, tf.int32)